Here’s the pattern I use when an EU software company starts billing serious volumes in South Africa, Kenya, and Nigeria. The goal isn’t exotic fintech—it’s predictable cash: let customers pay on local rails in their currency, keep refunds fast and auditable, and do FX at the holding with rules instead of vibes. When you wire it this way, approvals climb, reconciliation gets boring (the good kind), and your CFO stops discovering FX slippage at month-end.
What was breaking (and why it was expensive)
Before the redesign, the vendor invoiced in EUR or USD and told African customers to wire cross-border. That created the usual mess:
- Slow, short, and opaque: SWIFT hops shaved fees and days, broke remittance data, and spawned “missing payment” tickets.
- Refund drag: credits back to customers wandered through correspondent chains, turning service hiccups into relationship damage.
- FX roulette: ad-hoc conversions and deal-by-deal pricing meant spreads nobody could measure.
- Compliance anxiety: over-collection in hard currency, under-documented inflows, and refund exceptions made audits slow and defensive.
The fix was not “more payment methods.” It was local first for customers, policy first for treasury.
North star in one line
Let buyers pay on domestic rails in ZAR/KES/NGN with structured references and fast refunds; hold currency buffers in-country; convert at the holding on time/threshold sweeps; evidence everything.
Rails that actually clear (and refund)
Each market brings a reliable domestic backbone. You don’t need twenty options—you need two that never embarrass you:
- South Africa (ZAR)
- Domestic bank transfers via the national clearing systems; for small/urgent cases, instant retail rails (where enabled) shorten the last mile.
- Local card acquiring for SMB cohorts that insist on cards; keep descriptors recognizable and settlement domestic.
- Kenya (KES)
- Account-to-account transfers (including real-time interbank rails) for enterprises.
- Mobile money for SMB/retail-adjacent SKUs where adoption is dominant. Set strict refund mirroring to the original wallet or bank.
- Nigeria (NGN)
- Instant interbank transfers (the mainstream way businesses pay) with strong payer references; cards as a backstop for smaller tickets.
Across all three, we issued payer-bound virtual accounts or structured references so reconciliation became event-driven: when funds land, a webhook closes the invoice—no inbox archaeology.
Entity and cash model: keep customers local, keep FX centralized
You can run this with a partner of record or with light subsidiaries; the principles are the same:
- Local collection accounts per country in ZAR/KES/NGN for receivables and refunds.
- Working buffers sized to cover 7–10 days of outflows (refunds, local opex, taxes) so agents never see “insufficient balance to refund.”
- Intercompany ledger that treats each country unit as a cash pool operator for the holding.
- Holding-level FX: convert pooled ZAR/KES/NGN surpluses to EUR on threshold & calendar rules (e.g., when a pool tops its ceiling or ahead of payroll/AP). No per-invoice conversions.
This split keeps customer experience local and treasury decisions behind the curtain.
Pricing, contracts, and invoicing that don’t trip AP
- Native-currency invoicing: ZAR in South Africa, KES in Kenya, NGN in Nigeria—plus the buyer’s tax fields, PO numbers, and exact reference formats AP systems expect.
- Renewal language that explains method priority (bank transfer first, card as fallback, wallet where relevant) and refund timing in local time.
- Credit notes that mirror the original invoice for partial refunds or downgrades and attach to the same customer key and reference scheme.
Consistency gets you paid; creativity gets you tickets.
Refunds with clocks, not guesses
Refund policy is where trust is won:
- Return to source by default (same bank account or wallet).
- SLA you can say out loud: “Approved today → credited next business day” for domestic rails during banking windows; cards follow scheme timing.
- Pre-funded refund buckets inside each local pool so finance never blocks an agent.
Tie refunds to the original payment object in your ledger; revocations or license changes should post before the money moves back.
Risk, AML, and “boring but sharp” controls
- Name/account lock: first refund or payout can only go to a name-matched beneficiary; changes require out-of-band confirmation and a cooling-off period.
- Velocity caps: conservative refund and payout limits for new accounts; warm with clean behavior.
- Anomaly alerts: rapid pay-in/pay-out loops, third-country funding, or unusual mobile-money fragmentation raise a case.
- Sanctions & PEP checks at onboarding and periodically for corporate buyers and large tickets.
These controls stay invisible until needed—and then they save a bad afternoon.
Treasury and FX: measure spread, don’t feel it
At the holding, we ran three simple rules:
- Floors & ceilings per currency pool (ZAR/KES/NGN). Sweep the excess to EUR; top up below the floor.
- Benchmarked execution: every conversion logs achieved rate vs a neutral benchmark so realized spread is a number on a dashboard, not a debate.
- Light hedging only when balance duration exceeds policy comfort (e.g., big seasonality or committed opex).
No mid-transaction currency flips. Buyers never see your FX.
Reconciliation you can close on day two
Two decisions drove >95% auto-match:
- Virtual accounts / strict references per payer or per invoice so inbound credits self-identify.
- Evented accounting: “funds received,” “refund executed,” “conversion done,” “sweep sent” hit the ERP with immutable IDs and timestamps. The rest funnels to a tiny exceptions lane (partials, duplicates, late credits).
Month-end turned from archaeology into a checklist.
Results after three billing cycles
- Time-to-clear: same-day for most domestic transfers; minutes on instant rails and mobile money during banking hours.
- Unit economics: improved by triple-digit basis points versus cross-border wires and international cards as the default.
- Refund tickets: down sharply once agents could quote value dates in local time and hit them.
- Auto-recon: >95% with virtual references; exceptions were true edge cases.
- FX realized spread: measurable and stable, enabling vendor negotiation on facts.
Different mixes, same pattern.
Build order you can actually ship this quarter
Weeks 1–2
- Stand up local collection accounts in ZAR, KES, NGN (or through a regulated partner).
- Issue virtual references / virtual accounts. Publish a one-page “How to pay” with screenshots of popular bank apps/portals.
Weeks 3–4
- Wire webhooks into ERP; auto-close invoices on receipt.
- Create refund buckets with “next business day” SLA and source-account default.
Weeks 5–6
- Implement threshold-based sweeps to EUR with benchmark reporting.
- Expose FX dashboard (achieved vs benchmark) to treasury/CFO.
Weeks 7–8
- Ship credit-note workflow for downgrades/partial refunds.
- Add name/account lock, velocity caps, and anomaly alerts.
Weeks 9–10
- Tune method ordering by cohort (enterprise vs SMB; bank vs wallet), and add local card acquiring only where it improves approval without hurting refunds.
Ship corridors, not universes. Make it boring; boring scales.
Mistakes you don’t need to repeat
- Stuffing checkout with methods you can’t refund back to.
- Letting free-text references sneak into bank instructions “just once.”
- Converting every inflow the second it lands.
- Promising “instant refunds” without pre-funded local buckets.
You’ll pay for each of these—usually at month-end.